This paper is concerned with the development of a real-time obstacle avoidance system for an autonomous wheelchair using stereoscopic cameras by severely disabled people. Based on the left and right images captured from stereoscopic cameras mounted on the wheelchair, the optimal disparity is computed using the Sum of Absolute Differences (SAD) correlation method. From this disparity, a 3D depth map is constructed based on a geometric projection algorithm. A 2D map converted from this 3D map can then be employed to provide an effective obstacle avoidance strategy for this wheelchair. Experiment results obtained in a practical environment show the effectiveness of this real-time implementation.